- Devel - libvirt List Archives

[libvirt] Question about migration requirement
by Bruce Rogers 05 Jun '10

05 Jun '10

I've noticed that there seems to be some additional requirements placed on a migration, as compared to just remote management via libvirt. For example, when I try to do the following: virsh migrate running_vm_name qemu+ssh://1.2.3.4/system there seems to be a requirment that both the source and target hosts must resolve 1.2.3.4 the same way as far as naming services is concerned. Without being able to resolve to the same name on both the source and target, the migration will fail. Note that I can ssh to the other machine just fine using the ip address, or even do the following without problem: virsh -c qemu+ssh://1.2.3.4/system <some-command> Is this additional <requirement> for migration intentional (some security issue, etc), or is it just an artifact of implementation, which could be fixed to not be needlessly restrictive. Bruce

2 1

[libvirt] [PATCH] autobuild.sh: avoid bashism
by Eric Blake 05 Jun '10

05 Jun '10

* autobuild.sh: Replace 'set -o pipefail' with POSIX alternative. Reported by Matthias Bolte. --- On IRC, Matthias pointed out that his testing of my recent autobuild.sh patch triggered a dash failure. Unfortunately, the checkbashisms script is not smart enough to realize that 'set -o pipefail' is a bashism, yet it also does not provide a bug reporting address in checkbashisms --help :( autobuild.sh | 12 +++++++++--- 1 files changed, 9 insertions(+), 3 deletions(-) diff --git a/autobuild.sh b/autobuild.sh index 4de5af8..66ba132 100755 --- a/autobuild.sh +++ b/autobuild.sh @@ -35,9 +35,15 @@ esac make make install -set -o pipefail -make check 2>&1 | tee "$RESULTS" -make syntax-check 2>&1 | tee -a "$RESULTS" +# set -o pipefail is a bashism; this use of exec is the POSIX alternative +exec 3>&1 +st=$( + exec 4>&1 >&3 + { make check syntax-check 2>&1; echo $? >&4; } \ + | tee "$RESULTS" +) +exec 3>&- +test $st = 0 test -x /usr/bin/lcov && make cov rm -f *.tar.gz -- 1.7.0.1

2 2

[libvirt] Xen 4.0 and DPCI key in sexpr
by Matthias Bolte 05 Jun '10

05 Jun '10

This blog post [1] explains how to apply some OpenSuse patches [2] (especially xen-domctl-ver7.patch) to the libvirt 0.8.1 Debian package in order to get Xen 4.0 support. On IRC jonnyt reported problems with PCI passthrough in such a setup. Debugging and XenD hacking revealed that XenD 4.0 extended the sexpr format for PCI devices from (device (pci (dev (domain 0x0000)(bus 0x02)(slot 0x00)(func 0x0)))) to this, including a new key part (device (pci (dev (domain 0x0000)(bus 0x02)(slot 0x00)(func 0x0)(key <whatever-value-xend-expects-here>)))) The Xen upstream commit [3] that added this doesn't explain much about it. I came up with the attached prove-of-concept patch that just hardcodes (key 0x0) for XenD 4.0. On IRC jonnyt confirmed that this patch fixes the immediate error for him: error : xend_post:434 : POST operation failed: xend_post: error from xen daemon: (xend.err "Error creating domain: 'key'") Matthias [1] http://spblinux.de/blog/2010/05/xen-4-0-with-libvirt-0-8-1-replace-xm-new-b… [2] https://build.opensuse.org/stage/package/files?package=libvirt&project=Virt… [3] http://lists.xensource.com/archives/html/xen-changelog/2009-06/msg00298.html

2 3

[libvirt] Break up virsh man page into smaller sections?
by Justin Clift 04 Jun '10

04 Jun '10

Hi all, Looking at the virsh man page (tools/virsh.pod), it's getting pretty unwieldy and hard to grok for end users. Wondering if it's worth breaking it up into sections? Maybe something similar to how git structures their man pages? ie. man virsh Giving an overview, the categories DEVICE, DOMAIN, NETWORKING, STORAGE, (etc), and listing the more detailed man pages) man virsh-console (initially a copy of the existing virsh console man page contents.) virsh console domain-id Connect the virtual serial console for the guest. man virsh-create (initially a copy of the existing virsh create man page contents.) virsh create FILE Create a domain from an XML <file>. An easy way to create the XML <file> is to use the dumpxml command to obtain the definition of a pre-existing guest. Example # virsh dumpxml <domain-id> > domain.xml edit domain.xml virsh \ create < domain.xml <etc> ??? Regards and best wishes, Justin Clift -- Salasaga - Open Source eLearning IDE http://www.salasaga.org

2 2

[libvirt] [PATCHv3 0/2] build: automate VPATH testing
by Eric Blake 04 Jun '10

04 Jun '10

Picking up on a patch series that I had started a while ago: https://www.redhat.com/archives/libvir-list/2010-April/msg01259.html I finally have all the pieces in place for a successful run of ./autobuild.sh with a VPATH build. Now, we just have to remember to run ./autobuild.sh more frequently. Eric Blake (2): build: fix VPATH build of docs autobuild.sh: use VPATH build .gitignore | 1 + autobuild.sh | 10 +++++----- docs/Makefile.am | 46 +++++++++++++++++++++++++++------------------- docs/apibuild.py | 18 +++++++++++++----- 4 files changed, 46 insertions(+), 29 deletions(-)

2 3

[libvirt] [PATCH 0/1] Fix nodeDeviceDestroy return value
by David Allan 03 Jun '10

03 Jun '10

Set nodeDeviceDestroy's return value correctly in failure cases and clarify the error message generated when the parent device is not vport capable. David Allan (1): Fix device destroy return value src/conf/node_device_conf.c | 4 ++-- src/node_device/node_device_driver.c | 3 ++- 2 files changed, 4 insertions(+), 3 deletions(-)

4 4

[libvirt] [PATCH 0/1] Update nodedev scsi_host data
by David Allan 03 Jun '10

03 Jun '10

While investigating https://bugzilla.redhat.com/show_bug.cgi?id=597998 I noticed that my nodedev-create operations were timing out more often than not. This patch works around a race between udev generating a new device event and the corresponding sysfs entries being populated. This behavior has appeared before, but previously it was limited to incorrect WWNs; now no FC capability appears at all until the nodedev entry is refreshed. David Allan (1): Update nodedev scsi_host data before use src/node_device/node_device_driver.c | 19 +++---------------- 1 files changed, 3 insertions(+), 16 deletions(-)

4 4

[libvirt] [PATCH] Ensure UNIX domain sockets are removed on daemon shutdown
by Daniel P. Berrange 03 Jun '10

03 Jun '10

When libvirtd exits it is leaving UNIX domain sockets on the filesystem. These need to be removed. The qemudInitPaths() method has signficant code churn to switch from using a pre-allocated buffer on the stack, to dynamically allocating on the heap. * daemon/libvirtd.c, daemon/libvirtd.h: Store a reference to the UNIX domain socket path and unlink it on shutdown --- daemon/libvirtd.c | 127 ++++++++++++++++++++++++++++------------------------ daemon/libvirtd.h | 1 + 2 files changed, 69 insertions(+), 59 deletions(-) diff --git a/daemon/libvirtd.c b/daemon/libvirtd.c index e86f78d..381b2df 100644 --- a/daemon/libvirtd.c +++ b/daemon/libvirtd.c @@ -533,7 +533,7 @@ static int qemudWritePidFile(const char *pidFile) { } static int qemudListenUnix(struct qemud_server *server, - const char *path, int readonly, int auth) { + char *path, int readonly, int auth) { struct qemud_socket *sock; struct sockaddr_un addr; mode_t oldmask; @@ -549,6 +549,7 @@ static int qemudListenUnix(struct qemud_server *server, sock->port = -1; sock->type = QEMUD_SOCK_TYPE_UNIX; sock->auth = auth; + sock->path = path; if ((sock->fd = socket(PF_UNIX, SOCK_STREAM, 0)) < 0) { VIR_ERROR(_("Failed to create socket: %s"), @@ -743,17 +744,30 @@ cleanup: } static int qemudInitPaths(struct qemud_server *server, - char *sockname, - char *roSockname, - int maxlen) + char **sockname, + char **roSockname) { - char *sock_dir; - char *dir_prefix = NULL; - int ret = -1; + char *base_dir_prefix = NULL; char *sock_dir_prefix = NULL; + int ret = -1; + + /* The base_dir_prefix is the base under which all libvirtd + * files live */ + if (server->privileged) { + if (!(base_dir_prefix = strdup (LOCAL_STATE_DIR))) + goto no_memory; + } else { + uid_t uid = geteuid(); + if (!(base_dir_prefix = virGetUserDirectory(uid))) + goto cleanup; + } + /* The unix_sock_dir is the location under which all + * unix domain sockets live */ if (unix_sock_dir) { - sock_dir = unix_sock_dir; + if (!(sock_dir_prefix = strdup(unix_sock_dir))) + goto no_memory; + /* Change the group ownership of /var/run/libvirt to unix_sock_gid */ if (server->privileged) { if (chown(unix_sock_dir, -1, unix_sock_gid) < 0) @@ -761,68 +775,53 @@ static int qemudInitPaths(struct qemud_server *server, unix_sock_dir); } } else { - sock_dir = sockname; if (server->privileged) { - dir_prefix = strdup (LOCAL_STATE_DIR); - if (dir_prefix == NULL) { - virReportOOMError(); - goto cleanup; - } - if (snprintf (sock_dir, maxlen, "%s/run/libvirt", - dir_prefix) >= maxlen) - goto snprintf_error; + if (virAsprintf(&sock_dir_prefix, "%s/run/libvirt", + base_dir_prefix) < 0) + goto no_memory; } else { - uid_t uid = geteuid(); - dir_prefix = virGetUserDirectory(uid); - if (dir_prefix == NULL) { - /* Do not diagnose here; virGetUserDirectory does that. */ - goto snprintf_error; - } - - if (snprintf(sock_dir, maxlen, "%s/.libvirt", dir_prefix) >= maxlen) - goto snprintf_error; + if (virAsprintf(&sock_dir_prefix, "%s/.libvirt", + base_dir_prefix) < 0) + goto no_memory; } } - sock_dir_prefix = strdup (sock_dir); - if (!sock_dir_prefix) { - virReportOOMError(); - goto cleanup; - } - if (server->privileged) { - if (snprintf (sockname, maxlen, "%s/libvirt-sock", - sock_dir_prefix) >= maxlen - || (snprintf (roSockname, maxlen, "%s/libvirt-sock-ro", - sock_dir_prefix) >= maxlen)) - goto snprintf_error; - unlink(sockname); - unlink(roSockname); + if (virAsprintf(sockname, "%s/libvirt-sock", + sock_dir_prefix) < 0) + goto no_memory; + if (virAsprintf(roSockname, "%s/libvirt-sock-ro", + sock_dir_prefix) < 0) + goto no_memory; + unlink(*sockname); + unlink(*roSockname); } else { - if (snprintf(sockname, maxlen, "@%s/libvirt-sock", - sock_dir_prefix) >= maxlen) - goto snprintf_error; + if (virAsprintf(sockname, "@%s/libvirt-sock", + sock_dir_prefix) < 0) + goto no_memory; + /* There is no RO socket in unprivileged mode, + * since the user always has full RW access + * to their private instance */ } if (server->privileged) { - if (!(server->logDir = strdup (LOCAL_STATE_DIR "/log/libvirt"))) - virReportOOMError(); + if (virAsprintf(&server->logDir, "%s/log/libvirt", + base_dir_prefix) < 0) + goto no_memory; } else { - if (virAsprintf(&server->logDir, "%s/.libvirt/log", dir_prefix) < 0) - virReportOOMError(); + if (virAsprintf(&server->logDir, "%s/.libvirt/log", + base_dir_prefix) < 0) + goto no_memory; } - if (server->logDir == NULL) - goto cleanup; - ret = 0; - snprintf_error: - if (ret) - VIR_ERROR0(_("Resulting path too long for buffer in qemudInitPaths()")); +no_memory: + if (ret != 0) + virReportOOMError(); cleanup: - VIR_FREE(dir_prefix); + VIR_FREE(base_dir_prefix); VIR_FREE(sock_dir_prefix); return ret; } @@ -931,22 +930,22 @@ static struct qemud_server *qemudInitialize(void) { } static int qemudNetworkInit(struct qemud_server *server) { - char sockname[PATH_MAX]; - char roSockname[PATH_MAX]; + char *sockname = NULL; + char *roSockname = NULL; #if HAVE_SASL int err; #endif /* HAVE_SASL */ - roSockname[0] = '\0'; - - if (qemudInitPaths(server, sockname, roSockname, PATH_MAX) < 0) + if (qemudInitPaths(server, &sockname, &roSockname) < 0) goto cleanup; if (qemudListenUnix(server, sockname, 0, auth_unix_rw) < 0) goto cleanup; + sockname = NULL; - if (roSockname[0] != '\0' && qemudListenUnix(server, roSockname, 1, auth_unix_ro) < 0) + if (roSockname != NULL && qemudListenUnix(server, roSockname, 1, auth_unix_ro) < 0) goto cleanup; + roSockname = NULL; #if HAVE_SASL if (auth_unix_rw == REMOTE_AUTH_SASL || @@ -1057,6 +1056,8 @@ static int qemudNetworkInit(struct qemud_server *server) { return 0; cleanup: + VIR_FREE(sockname); + VIR_FREE(roSockname); return -1; } @@ -1080,6 +1081,7 @@ static int qemudNetworkEnable(struct qemud_server *server) { return 0; } + static gnutls_session_t remoteInitializeTLSSession (void) { @@ -2422,6 +2424,13 @@ static void qemudCleanup(struct qemud_server *server) { if (sock->watch) virEventRemoveHandleImpl(sock->watch); close(sock->fd); + + /* Unlink unix domain sockets which are not in + * the abstract namespace */ + if (sock->path && + sock->path[0] != '@') + unlink(sock->path); + VIR_FREE(sock); sock = next; } diff --git a/daemon/libvirtd.h b/daemon/libvirtd.h index d292681..4d8e7e2 100644 --- a/daemon/libvirtd.h +++ b/daemon/libvirtd.h @@ -233,6 +233,7 @@ struct qemud_client { struct qemud_socket { + char *path; int fd; int watch; int readonly; -- 1.6.6.1

2 1

[libvirt] [PATCH] Disable use of 'reason' field in block IO event in QEMU
by Daniel P. Berrange 03 Jun '10

03 Jun '10

QEMU upstream decided against adding a 'reason' field to the block IO event in QMP. Disable this code to remove a annoying warning message. It will be renabled when the error string reason is re-introduced in QEMU --- src/qemu/qemu_monitor_json.c | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-) diff --git a/src/qemu/qemu_monitor_json.c b/src/qemu/qemu_monitor_json.c index f56bcdc..e89ad43 100644 --- a/src/qemu/qemu_monitor_json.c +++ b/src/qemu/qemu_monitor_json.c @@ -589,10 +589,14 @@ static void qemuMonitorJSONHandleIOError(qemuMonitorPtr mon, virJSONValuePtr dat VIR_WARN0("missing device in disk io error event"); } +#if 0 if ((reason = virJSONValueObjectGetString(data, "reason")) == NULL) { VIR_WARN0("missing reason in disk io error event"); reason = ""; } +#else + reason = ""; +#endif if ((actionID = qemuMonitorIOErrorActionTypeFromString(action)) < 0) { VIR_WARN("unknown disk io error action '%s'", action); -- 1.6.6.1

2 1

[libvirt] Draft of pre migration checks framework
by Paolo Smiraglia 03 Jun '10

03 Jun '10

Hi guys, I'm working on a pre migration checks implementation. The patch files attached provide a framework to execute pre-migration-checks before the domain migration. ================================================================ 1 - FRAMEWORK OVERVIEW ================================================================ Pre-Migration-Checks are enabled by adding a "--pmc" option in "virsh migrate" command $ virsh migrate --pmc ... In this way, before the execution of function virDomainMigrate(), an array of "checks" are performed. Return value of every check set if the migration will continue or not. ================================================================ 2 - FRAMEWORK STRUCTURE ================================================================ Every check is defined in the file libvirt/src/util/pmc.c Check list is defined by object "chklist". This is a struct defined as static virPMCCheckList chklist = { .count = 2, .check = {&chk1,&chk2,NULL} }; where "count" define the current array size, and "check" is an array of pointer to virPMCCheck object. virPMCCheck is a stuct defined as struct _virPMCCheck { const char *name; virPMCCheckRun run; }; where "name" is the name of the check needed for debugging messages, and "run" is a pointer to check function. ================================================================ 3 - HOW TO ADD NEW CHECK ================================================================ Adding new check is very simple! You have to define a check funcion as static int pmcCheckFooCheck(virDomainPtr domain, virConnectPtr dconn) { /* do something */ return CHECK_SUCCESS } then you have to define a "check-hook" as static virPMCCheck chk3 = { .name = "this is fooCheck", .run = pmcCheckFooCheck }; and update the object "chklist" static virPMCCheckList chklist = { .count = 3, /* previous value was 2 */ .check = {&chk1,&chk2,&chk3,NULL} /* previous content not * include a pointer to * chk3 object */ }; ================================================================ 4 - POSSIBLE ENHANCHEMENT ================================================================ 1. Adding infos in _virPMCCheck struct about check security level. Example: if a check with level CRITICAL return CHECK_FAIL, migration process is stopped, 2. Implementing framework not as util, but as driver. For example this can permit to define new checks by xml file. 3. Define external checks loadable by shared library. ================================================================ 5 - PATCH FILE DESCRIPTION ================================================================ file: datatypes.h.patch desc: define new type virPMCCheckRun define struct _virPMCCheck define struct _virPMCCheckList file: libvirt.c.patch desc: impement public API virDomainMigratePMC() file: libvirt.h.in.patch desc: impement public API virDomainMigratePMC() prototype define type virPMCCheck define type virPMCCheckPtr define type virPMCCheckList define type virPMCCheckListPtr define macro PMC_LIST_SIZE file: libvirt_public.syms desc: export public API virDomainMigratePMC() file: pmc.c.patch, pmc.h.patch desc: define and implements pre migration checks and some auxiliary functions. file: virsh.c.patch desc: add option "pmc" to virsh migrate command ================================================================ 6 - NOTES ================================================================ I start to implement this framework on git sources, but the latest sources produce a regress on my code. For this reason I applied my code on latest stable release of libvirt (0.8.1). What do you think about that? Is this a good approach to implement pre migration checks? Have you a suggestions? There is a possibility to include this patch in future libvirt distributions? Waiting for feedback... Paolo Smiraglia -- PAOLO SMIRAGLIA http://portale.isf.polito.it/paolo-smiraglia

2 2